Fast FCN for Medical Hyperspectral Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CNN-based medical hyperspectral image (MHSI) classification methods suffer from low computational efficiency due to redundant computation in overlapping regions between image patches, and they fail to fully utilize spatial contextual information.
Innovation Solution
The proposed method employs a fast fully convolutional network (FCN) for MHSI classification, which includes preprocessing and sampling of MHSI to obtain a training sample set, followed by training using an encoder-decoder-based FCN. This approach avoids redundant computation by using a Convolutional Attention Module (CBAM) and lateral connection-based semantic-spatial fusion to maximize the utilization of global spatial information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If patch-based local learning framework is used for MHSI classification, then local features can be extracted, but redundant computation occurs due to overlapping of image patches of adjacent pixels and operation speed is limited
Solution Approach 1:
The patent merges multiple overlapping patch processing operations into a single unified FCN model that processes the entire image at once. The encoder-decoder architecture with CBAM modules combines spatial and spectral information globally, eliminating redundant computations across overlapping patches while maintaining the ability to capture local features through convolutional operations.
2Measurement precision
If spatial-spectral information fusion is used for classification, then classification precision can be improved, but computation time increases due to the spatial-spectral information fusion process
Solution Approach 1:
The patent implements continuous spatial-spectral feature extraction through the encoder-decoder architecture where CBAM (Convolutional Block Attention Module) operations are integrated throughout the network. The attention mechanisms continuously refine spatial and spectral features in a single forward pass, eliminating the need for separate processing stages and reducing overall computation time while maintaining high classification precision.
3Productivity
If image patches of small size are used for processing, then processing speed can be improved, but only local features can be extracted and classification performance is limited
Solution Approach 1:
The patent transitions from processing small 2D image patches to processing the entire 3D hyperspectral image volume simultaneously using the FCN architecture. The encoder-decoder structure with CBAM modules enables the model to capture global spatial context and spectral information across the entire image, effectively adding a global dimension to feature extraction while maintaining processing efficiency through optimized convolutional operations.
Data Source
AI summary
The present disclosure provides a medical hyperspectral image (MHSI) classification method based on a fast fully convolutional network (FCN), and relates to the technical field of MHSIs. The MHSI classification method includes: preprocessing and sampling an MHSI to obtain a training sample set; inputting the training sample set into an encoder-decoder-based FCN to train the MHSI; and inputting a to-be-classified pixel of the MHSI into a trained encoder-decoder-based FCN to obtain a classification result. The present disclosure provides the MHSI classification method based on a fast FCN. In order to resolve problems of low efficiency and insufficient performance of an existing MHSI classification method, the present disclosure designs a classification method based on the fast FCN, which avoids redundant computation in an overlapping region between image patches, greatly improving an inference speed.